Adaptive multipath equalization

Abstract

Method and apparatus for using a plurality of correlators to improve an estimate of direct signal arrival time by identifying features of a correlation function at and adjacent to the correlation peak. In a first embodiment, the errors in location of the center point of a correlation function r(τ), formed by the incoming composite signal and a stored copy of the expected signal, are assumed to be strongly correlated for narrow sample spacing and wide sample spacing of the correlation function. In a second embodiment, multipath signal strengths and phases are estimated, using multiple sampling of the correlation function r(τ). This approach assumes that path delays of the direct signal and of the multipath signals can be determined separately. Path delays can be determined by any of at least three approaches: (1) identification of slope transition points in the correlation function; (2) Cepstrum processing of the received signal, using fourier transform analysis; and (3) use of a grid of time points on the correlation function domain, and identification of time values, associated with certain solution parameters of the least mean squares analysis that have the largest absolute values, as times of arrival of the direct and multipath signals. Separate identification of multipath time delays reduces the least mean squares analysis to a linear problem. A modified signal is constructed, with the multipath signal(s) approximately removed from the incoming composite signal. This modified signal allows a better estimate of the arrival time of the direct signal.

Description

FIELD OF THE INVENTION

This invention relates to adaptive equalization of, and compensation for, multipath electromagnetic wave signals received and used for navigation and similar purposes.

BACKGROUND OF THE INVENTION

Many sophisticated navigation systems rely upon receipt of a sequence of electromagnetic wave signals, such as radio wave signals, from signal sources that may be tens of miles to thousands of miles from the receiver. The signals may be reflected or otherwise affected and thus arrive at the receiver as two or more multipath signals displaced in time from each other. Ideally, the receiver will either severely discriminate against all but one of these replicas; or the receiver will equalize and/or compensate for the time delays and shape distortions present in the composite signal.

Several workers have proposed approaches for reducing or removing multipath signal errors. Lakatos, in U.S. Pat. No. 3,537,008, distinguishes between a direct or "specular" path from transmitter to receiver and one or more "echo" paths that provide multipath signals for that transmission. A modulated pilot signal is transmitted together with a message signal, and arrival of the known modulated signal is continuously monitored at the receiver. The modulated signal is of finite length and acts as a timing signal to open and close a gate at the receiver. The contemporaneously transmitted direct signal is admitted through the gate, but later arriving echo signals are not admitted. This system uses two parallel signal processing paths at the receiver, one path containing a phase locked loop for recognition of the modulated signal. If an echo signal arrives before the receiver gate closes, a portion of the echo signal will also be admitted and treated as part of the direct message signal.

Decomposition of the arriving signal into selected contiguous but nonoverlapping frequency intervals to reduce multipath signal error is disclosed in U.S. Pat. No. 3,542,954, issued to Flanagan. The transmitted direct signal and any echo signals are received and normalized at a plurality of spaced apart transducers, at each of which this frequency decomposition occurs. Among the transducers, the maximum signal in each frequency interval is chosen, and a composite direct signal at the intended receiver is constructed from this maximum signal set.

U.S. Pat. No. 3,566,035, issued to Noll et al, discloses use of the "complex cepstrum" (Fourier transform of the logarithm of the power spectrum plus phase angle) of short time interval, partly overlapping segments of a transmitted signal to determine periodicity or aperiodicity of a received signal. Flanagan discloses application of the cepstrum function to multipath signal suppression, in U.S. Pat. No. 3,662,108. The complex cepstrum of a received composite signal is formed, and the inverse Fourier transform of the real part of the cepstrum is computed. This inverse transform signal is clipped about a selected center to remove some distortion components, and the remaining "mountain top" signal has a sequence of maxima at time values corresponding to arrival of the desired signal and undesired multipath replicas at the receiver. A fast Fourier transform technique for computing cepstrum functions from received signals is disclosed by Dieterich et al in U.S. Pat. No. 5,065,242, for use in removing ghost images in received television signals.

Menard discloses a multipath time delay and correlation bandwidth analyzer system in U.S. Pat. No. 3,596,182. A correlation function is formed between a signal received at the receiver and a selected reference signal generated at the receiver. If the received signal contains strong multipath contributions, displaced in time from each other, the correlation function will contain two or more correlation pulses or maxima, also displaced in time. The reference signal can be a time delayed replica of the received signal.

U.S. Pat. No. 3,605,018, issued to Coviello, discloses a nonlinear signal processing system useful for suppressing strong multipath signals in a quaternary phase, spread spectrum communications system. A received composite signal, containing the desired signal and a superimposed strong multipath signal, is passed through a circuit containing a signal envelope detector and a signal averager, arranged in parallel. The envelope detector and signal averager output signals are applied to two terminals of a differential amplifier. The diff amp output signal is an estimate of the desired signal.

A system for quantitatively measuring multipath signal distortion is disclosed by Close in U.S. Pat. No. 3,869,673. A local oscillator, mixer and amplitude limiter are applied to a received signal to provide a first signal representing frequency variation and a second signal representing amplitude variation of the received signal. A correlation signal formed from the product of these first and second signals is a quantitative measure of multipath distortion present in the received signal.

Costas discloses apparatus for minimization of distortion in signals containing multipath signals and Doppler shift effects in U.S. Pat. No. 4,349,915. Signal arrival delay, such as produced by multipath signals, is determined by examining a selected frequency contribution of spaced apart pulses that arrive at the receiver.

Gutleber, in U.S. Pat. No. 4,457,007, discloses a multipath signal interference reduction system in which a plurality of replicas of the received signal, with different time delays, amplitudes and signal widths, are sequentially subtracted from the arriving composite signal (desired signal plus multipath signals) to produce a new composite signal with markedly reduced multipath signal contribution.

An adaptive multipath distortion equalizer, disclosed by Nossen in U.S. Pat. No. 4,669,091 uses simulation of the multipath distortion, based upon the composite signals received. After this multipath distortion signal is estimated, this distortion signal is subtracted from the composite signal received to produce an estimate of the desired signal. This approach uses time delay lines, amplitude scale factors and phase adjustments to obtain the estimated multipath distortion signal.

In U.S. Pat. No. 4,672,638, Taguchi et al disclose a multipath signal canceller that uses an envelope detector to determine the shape of a received composite pulse. This shape is compared with the shape of an undistorted pulse (earliest arriving pulse) stored in the receiver. The difference between these two shapes is used to construct a distortion-cancelling signal for signals that subsequently arrive at the receiver.

Rilling discloses a multipath signal reduction system, intended to compensate for multipath signals arriving from unknown directions, in U.S. Pat. No. 4,797,950. An adaptive array of spaced apart antennas is provided, with each antenna signal being processed identically by a band pass filter, a local oscillator, a signal mixer and a tapped time delay line. The processed signals are adaptively weighted by a feedback loop and added together to provide a signal with reduced multipath contributions. Any of eight classes of signal feedback equations can be used, depending upon the choice of form of an error term used for least squares minimization of a error signal. No reference signal is required for comparison in this system. An antenna pattern null is adaptively positioned in the (unknown) direction of the arriving, unwanted multipath signals.

Schilling, in U.S. Pat. No. 5,081,643, discloses multipath receiver apparatus that compares a time delayed replica of a transmitted with the received signal and chooses a signal propagation time delay that maximizes a correlation signal formed from the first two signals.

A method for multipath signal suppression in television receivers, disclosed by David Koo in U.S. Pat. No. 5,111,298, uses transmission of a test signal that is sampled and transformed to the frequency domain. Portions of the frequency spectrum are redistributed to other frequency ranges where the spectrum amplitude is nearly zero, and the redistributed spectrum is transformed back to the time domain to produce a processed signal where echo signal contributions are suppressed. Koo discloses another method for echo signal cancellation in television receivers in U.S. Pat. No. 5,121,211, again working with the frequency domain transform of a received test signal.

Chan et al, in U.S. Pat. No. 5,127,051, disclose an modem system that senses, and adaptively adjusts itself for, the presence of multipath signals in the transmission channel. A message is decomposed into a sequence of frames, each containing a first training packet, a data packet, and a second training packet. Each training packet contains a selected sequence of binary symbols and has a length that exceeds the expected temporal length of multipath dispersion in the channel. As each three-packet frame is received, the received first and second training packets are compared with the expected first and second training packets, symbol-by-symbol, to determine and compensate for the presence of any multipath dispersion in the channel.

In U.S. Pat. No. 5,142,551, Borth et al disclose a signal weighting equalizer for adaptively filtering multipath signals received by a single antenna. The system determines and assigns confidence levels on signal accuracy to different portions of a received signal and processes these portions according to the confidence levels assigned thereto. A received signal is passed through an equalizer, including a maximum likelihood sequence estimator, at the receiver to estimate the signal originally transmitted. A synthesized transmission channel generates and modifies a synthesized "re-transmitted" signal, based upon the equalized signal, that is compared with the signal actually received at the receiver to determine confidence levels for different portions of the received signal.

A multipath detector containing a differential amplifier that senses whether the signal strength is above a first level or is below a second (lower) level is disclosed by Nakamura et al in U.S. Pat. No. 5,159,708. If the signal strength is below the second level, the system concludes that a multipath signal is present. The signal threshold for presence of a multipath signal increases approximately linearly with an average electrical field strength of a received signal.

Cai et al disclose a digital signal equalization system that senses whether multipath dispersion is present in a received, data-modulated signal, in U.S. Pat. No. 5,164,959. A correlation signal is formed between the data-modulated signal and a similar, but non-data-modulated, signal generated at the receiver. If the correlation signal has two or more peaks, the system concludes that multipath dispersion is present and estimates the transmission channel impulse response. The system then determines a set of optimum weights for the direct and multipath signals to align these signals in phase and time to estimate the signal that would have been received through a non-dispersive channel.

Belcher et al, in U.S. Pat. No. 5,203,016, disclose a signal quality-dependent recursive integrator that compares the amplitude of an emphasized signal information component with a threshold amplitude to control gain in a feedback delay loop, for analyzing arriving signals containing multipath contributions. One or more time delayed versions of the received signal is recursively summed with the originally received signal, and the feedback loop gain varies inversely with the signal-to-noise ratio of the received signal.

Minamisono, in U.S. Pat. No. 5,218,359, discloses a post-processing adaptive antenna array system with independently controllable weights and phases assigned to each antenna. Correlation functions associated with different times and directions of signal arrival are computed. The antenna weights and phases are chosen to correspond to a direction of signal arrival that produces a cumulatively high received signal level, to suppress multipath signal interference.

In U.S. Pat. No. 5,241,317, issued to Howard, multipath signal reduction apparatus for radar signal analysis for range, altitude and elevation angle is disclosed. The apparatus receives sum and difference radar return signals from a target and logarithmically amplifies these signals. A largest peak amplitude signal is selected for each of the fading sum signals and the fading difference signals, and these selected signals are used to determine a ratio that is related to target elevation angle. This approach relies on identification of peak signal values and does not directly identify a multipath signal that may be present.

Ohsawa, in U.S. Pat. No. 5,303,264, discloses an adaptive equalizer for reducing signal distortion due to multipath interference and fading in a digital system. A "signal forget factor", dependent on the signal-to-noise ratio of an incoming signal, is applied to an automatic gain controller that receives the incoming signal. The equalizer varies the waveform with forget factor included, using a recursive least squares algorithm.

Several of these patents disclose examination of a correlation signal of, or form a related comparison of, an arriving signal with a selected reference signal generated at the receiver. These approaches usually require examination of a correlation signal for the presence of two or more local maxima or peaks, which are taken to indicate the presence of multipath signals. Here, however, interest centers on mitigating the presence of multipath signals that are delayed in time by less than the width of one digital symbol time, referred to as a chip width. These multipath signals do not necessarily manifest their presence by appearance of distinct correlation peaks or local maxima, but in other, more subtle ways. What is needed is an approach allowing more accurate estimation of the arrival time of the direct path signal by means of accurate quantitative evaluation of characteristics of the non-direct or multipath signals. Preferably, this approach should allow (near) real time identification and mitigation of the multipath signals so that the arrival time of the direct signal can be estimated with greatest accuracy.

SUMMARY OF THE INVENTION

The invention meets these needs by providing method and apparatus for using a plurality (preferably four or more) of correlators to identify detailed features of a correlation function at and adjacent to the correlation peak. In a first embodiment, the errors in location of the center point of a correlation function, formed by the received signal and a stored copy of the expected direct signal, are assumed to be strongly correlated for narrow sample spacing and wide sample spacing of the correlation function.

In a second embodiment, the multipath signal strengths and phases are estimated by a least mean squares analysis, using multiple sampling of a correlation function of an expected signal and an arriving composite signal that includes the direct signal and one or more multipath signals. This approach assumed that the times of arrival or path delays of the direct signal and the multipath signals are determined separately.

In one method of estimating these path delays, the invention forms an error function, which is a difference between correlation functions R(τ) of "early-arriving" p(t+τ_d) and "late-arriving" signals p(t-τ_d) with a reference signal p(t), which may be a pseudo-random noise code signal, to determine whether multipath signals are present and to adaptively determine appropriate weighting to remove or suppress such signals. These two signals are combined to form a single, non-coherent correlation function (complex phase not present) R(τ_d) that achieves a local maximum for a particular time translation. The number of slope transition points that appear in the correlation function R(τ_d) for τ_d ≠0 indicates the number of strong multipath signals present and their times of arrival relative to the direct or line-of-sight signal. The apparatus determines times τ_d where this slope changes quickly or abruptly. A slope transition point occurs wherever this slope changes by more than a selected amount.

In another method of estimating these paths delays, a complex cepstrum function, discussed in the Noll et al and Flanagan patents, is used to estimate the strength and arrival time of any multipath signals present.

Once the full set of path delays {τ_n }_n is determined, determination of the complex gain parameters associated with the direct and multipath signals becomes a linear problem, if a least squares approach or similar method is used for such determination. Thus, use of the slope transition approach or of the cepstrum processing approach decomposes the problem of determination of the parameter sets {τ_n }_n, {a_n }_n, and {θ_n }_n into two independent sub-problems, and the second of these sub-problems merely requires solution of an array of linear equations.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic view of a prior art signal receiver used for analyzing multipath signals.

FIG. 2 is a schematic view of receiver apparatus operated in the presence of multipath signals.

FIG. 3 is a graphical view of a correlation signal formed from receipt of a direct signal only.

FIG. 4 is a graphical view of a correlation signal that might be formed from receipt of a composite signal, including multipath signal(s).

FIG. 5 is a graphical view of a correlation function in the presence of one or more multipath signals, indicating the center point error that occurs with use of narrow spaced and wide spaced correlator samples.

FIG. 6 illustrates a prior art receiver.

FIG. 7 illustrates an embodiments of a receiver constructed according to the invention.

FIGS. 8A, 9A and 10A are graphical views of correlation signals formed from receipt of three different composite signals, and FIGS. 8B, 9B and 10B are graphical views of error signals formed from these respective correlation signals, according to one embodiment of the invention.

FIGS. 11 and 12 are graphical views of errors illustrate results of simulation of multipath signal-induced errors in center point position.

FIGS. 13A and 13B illustrate a cepstrum signal with no multipath signals present and in the presence of at least one multipath signal, respectively.

FIG. 14 illustrates apparatus suitable for removing one or more multipath signals from a signal using a cepstrum analysis approach.

FIGS. 15, 16 and 17 are flow charts illustrating procedures for three embodiments of the invention.

DESCRIPTION OF BEST MODE OF THE INVENTION

FIG. 1 illustrates a receiver used in the prior art. A received signal s(t) is represented by ##EQU1##

s₀ (t)=a₀ p(t-τ₀) exp[j(ωt+θ₀)],(2)

s_m (t)=a_m p(t-τ_m) exp[j(ωt+θ_m)] (m=1, . . . , N), (3)

where the first term s₀ (t) is the "direct" term and the sum S_M (t;N) of the N additional terms s_m (t) arises from N additional (multipath) signals (N≧1) that arrive at the receiver by other reflection and/or refraction. The parameters a_m, τ_m and θ_m (m≧0) represent the amplitude, time delay and phase, respectively, of the mth signal component and p(t) is a known function. The received signal s(t) is passed through a bandpass filter (BPF) 11 (optional) and is converted to a baseband signal by multiplication in a signal multiplier at 13 of this signal by a numerically controlled local oscillator signal, represented by exp[-j(ωt+θ*)] and produced by a local oscillator 15, and passage of the resulting signal through a low pass filter (LPF) 17, to produce a baseband signal ##EQU2## Here, θ* is the receiver's estimate of the phase of the original signal s(t).

It is assumed herein that the transmitted signal is digital and has a known carrier signal that is a sinusoidal signal with a frequency ω, modulated with a preselected binary signal p(t) with a prescribed chip frequency f_c. The selected binary signal may be, but need not be, a pseudo-random noise signal, for example, as used in a global positioning system or global navigation system, such as GPS or GLONASS.

Typically, a receiver will form a correlation function of the baseband signal s_bb (t) with two or three copies of the expected signal p(t-τ₀), suitably time shifted. If the local oscillator signal multiplication, low pass filtering and correlation functions are combined, the result is two or more correlation functions of the form ##EQU3## where τ* is the present estimate of the signal arrival time and τ_d is the separation time for two samples centered at the time τ*, and T is the integration time for the integration modules 25 and 27. The separation time τ_d is between 0.05 T_c and 0.2 T_c, and between 0.5 T_c and T_c, for narrow correlator spacing and wide correlator spacing, respectively, where T_c =1/f_c is the length of the time interval of one pulse, or "chip width". If the estimated arrival time τ* is within a time interval of length τ_d /2 of the true arrival time τ, the early signal x₁ and the late signal x₂ "straddle" the true signal arrival time τ, whose value is not yet known. The early and late code functions p[t-τ*±τ_d /2] are stored and/or generated by a code generator 19, and the early and late code functions are multiplied by the baseband signal s_bb (t) in two signal multipliers 21 and 23, respectively. The product signals produced by the multipliers 21 and 23 are passed to the integrator modules 25 and 27, where the input signals are integrated over a time interval [0,T] to form the correlation signals x₁ (τ*,τ_d) and τ₂ (τ*,τ_d). The difference of these two correlation signals is formed in a difference module 29, and the difference signal is passed to a code loop filter 31 that determines appropriate values of the variable τ* that "steer" the quantity x₁ (τ*,τ_d)-x₂ (τ*,τ_d) toward 0. The output signal (τ_d and τ*) from the code loop filter 31 is passed to the code generator module 19, which uses these input values to form the code functions p(t-τ*±τ_d /2).

The code generator 19 also produces a "prompt" signal p(t-τ*), which is multiplied in a multiplier module 33 by the original baseband signal s_bb (t) and is passed to a carrier loop filter 35 that develops an error signal. This error signal steers the frequency and phase of the local carrier oscillator. The output signal of the carrier loop filter 35 is received by the local oscillator 15 and controls the choice of LO frequency and/or LO phase θ*. The carrier feedback loop shown in FIG. 1 seeks to drive the imaginary part of the output signal of the carrier loop filter 35 to zero.

In FIG. 2, a signal p(t)exp[j(ωt+θ₀)] transmitted by a transmitter 41 is received by a receiver 43 as the sum of (1) a "direct" signal s₀ (t)=a₀ p(t-τ₀)exp[j(ωt+θ₀)] plus (2) one or more possibly distorted replicas s_m (t), called multipath signals, of the direct signal at arrival times t=t_m ≠τ₀ (m=1, 2, . . . , N), represented as a sum s(t)=s₀ (t)+S_M (t;N). One multipath signal component s₁ (t) may have been reflected from a ground plane 45 before that component arrives at the receiver 43. A second multipath signal component s₂ (t) may have been reflected by an object 47 spaced apart from the Earth before being received at the receiver 43.

If only additive white Gaussian noise is assumed to be present, the variance of the estimated time delay error τ-τ* is found to be

σ² (τ*)=E[(τ-τ*)² ]=T_c τ_d B_N /(C_p /N₀) (in sec²), (7)

where B_N is the noise bandwidth of the closed code phase loop, and C_p /N₀ is the ratio of carrier power to the noise spectral power density. Equation (7) indicates that, all other things being equal, a receiver with narrow correlator spacing τ_d will provide smaller variance, and thus better performance, than a receiver with large correlator spacing.

A choice of narrow correlator sample spacing τ_d <<T_c is also sometimes appropriate in a multipath signal environment, because the error in estimation of arrival time τ generally grows with the correlator spacing τ_d. Although narrow correlator sample spacing generally reduces the mean squared error for the estimated arrival time t, certain demanding applications require even smaller errors. Further, if the narrow correlator sample spacing is too small, where a correlation function has more than one local maximum or peak, the receiver can "lock" to the wrong peak by receiving correlation information only for very small spacings.

The invention uses four or more correlator sample values, or correlations, to obtain better performance where multipath signals are present. The invention relies upon one or more of four approaches: (1) the correlation function is sampled using narrow sample spacing (τ_d ≈0.05 T_c to 0.2 T_c) and using wide sample spacing (τ_d 0.5 T_c to T_c), with the additional data from the wide sample spacing being used to correct errors introduced by using only narrow sample spacing in a multipath environment; (2) least squares estimation of multipath signal strengths, wherein the correlation function is multiply sampled (typically 10-30 times over the correlator function width) and an algorithm provides an estimate of the multipath signal strengths, by minimizing the sum of certain square error residuals; (3) least squares estimation of multipath signal strength parameters, where path delays are identified by one or more times where the slope of a correlation function changes abruptly and substantially; and (4) least squares estimation of multipath signal strength parameters, where path delays are identified using cepstrum signal processing of the received signal s(t), manipulating the logarithm of the Fourier transform of s(t), followed by an inverse Fourier transformation. The multipath manifests its presence by a sequence of pulses, each corresponding to arrival of a distinct direct or multipath signal in this algorithm.

FIG. 3 illustrates a correlation function magnitude C(τ)=|R(τ)|, where R(τ) is defined by Eq. (6) and the baseband signal s_bb (t) is defined in Eq. (4). In FIG. 3, no multipath signals are present (S_M (t;N)=0 in Eq. (1)). The correlation function magnitude |R(τ)| is approximately symmetric, reaches a single maximum value at τ=τ₀ and decreases smoothly and strictly monotonically to zero as τ→τ₀ ±T_c, where f_c =1/T_c, where T_c is the chip width and f_c is the chipping rate.

FIG. 4 illustrates a representative correlation function magnitude C(τ)=|R(τ)| that might be generated for the total signal s(t) of Eq. (1), with direct and multipath signals present. Here, the correlation function has a maximum at τ=τ₀ as before and manifests a significant change in slope at τ=τ₀, at τ=τ₀ -0.7 T_c and at τ=τ₀ +0.35 T_c, and the slopes of the correlation function magnitude C(τ) for τ<τ₀ and for τ>τ₀ differ substantially from each other. At each time slope transition point τ=τ₀ +t_s in the domain of the correlation function, the slope changes abruptly. Each slope transition point τ=τ₀ +t_s potentially corresponds to a time of arrival τ=τ₀ +t_s of a distinct multipath contribution to the signal S_M (t;N) in Eqs. (1)-(3). By determining the time value of each slope transition point τ=τ₀ =t_s, one may identify the corresponding slope transition times τ_m =τ₀ +t_s,m and the integer N to be used to construct an approximation for the function s_bb (t) in Eq. (4) representing the baseband signal that arrives at the receiver 43 in FIG. 2. This analysis is discussed below.

FIG. 5 exhibits another representative correlation function magnitude C(τ)=|R(τ)| that might appear if multipath signals are present. FIG. 5 illustrates the results of: (1) wide correlator function sampling, with corresponding sampling points τ=τ₁,w and τ=τ₂,w, a midpoint τ=τ_w,avg =(τ₁,w +τ₂,w)/2, and an associated midpoint error Δτ_w =τ_w,avg -τ₀ ; (2) narrow correlator function sampling, with corresponding sampling points τ=τ₁,n and τ=τ₂,n, a midpoint τ=τ_n,avg =(τ₁,n +τ₂,n)/2, and an associated midpoint error Δτ_n =τ_n,avg -τ₀ ; and (3) a midpoint error difference Δτ_w -Δτ_n =Δτ_w,avg -Δτ_n,avg between the wide and narrow correlator function sampling values.

For relatively narrow correlation sample spacing (e.g., τ₂n -τ₁n =τ_d ≈T_c /20) in FIG. 5, the estimated narrow sample spacing midpoint τ_n,avg for this spacing is determined by requiring that the early and late correlator function amplitudes (or their absolute values) be equal,

|R(τ₁,n)|=|R(τ₂,n)|(8 )

for the particular choices τ₁n and τ₂n made. The midpoint error Δτ_n is relatively small. For relatively wide correlation sample spacing (e.g., τ₂w -τ₁w =τ_d ≈T_c /2) in FIG. 5, the early and late sample point amplitudes or magnitudes |R(τ₁,w)| and |R(τ₂,w)| are again required to be equal. The estimated wide sample spacing midpoint τ=τ_w,avg is displaced from the true center point τ=τ₀ by a larger amount Δτ_w =τ_w,avg -τ₀. However, as demonstrated later, the "wide" samples contain information that can be used to further reduce the error suffered by the "narrow" samples.

A multipath noise canceller 51, well known to those skilled in the art, is illustrated in one form in FIG. 6. The noise canceller 51 attempts to replicate the total multipath signal component S_M (t;N), for subsequent subtraction from the filtered signal s_f (t). The input signal s(t) is passed through a bandpass filter (BPF) 53, and the filtered input signal s_f (t) is received by a first input terminal of a difference module 55. A second input terminal of the difference module 55 receives a sum signal s_M (t;N) formed by a sum module 57, whose operation is described below. The output signal Δs(t)=s_f (t)-s_M (t;N)≈s₀ (t) (the direct signal) is passed to a first input terminal of a correlator module 59 that includes a memory unit 61. The filtered signal s_f (t) is received at a second input terminal of the correlator module 59. The memory unit 61 contains a known shape function p(t) that characterizes the signal transmitted by the transmitter 11 in FIG. 2, which is provided as a first input signal to a multiplier module 63, whose second input signal is a local oscillator signal exp[-j(ω_IF t+θ*)], where the intermediate frequency ω_IF and the phase angle θ are selected from other considerations.

The output signal from the multiplier module 63 is of the form p(t)exp[-j(ω_IF t+θ*)], and this output signal is transmitted along a tapped time delay line 65 having N time delay taps (N≧1) corresponding to time delays τ1, τ2, . . . , τN, where these individual time delays are fixed or chosen by an algorithm module 67 that receives an input signal from the correlator module 59. At a given time delay tap (m), the output signal is p(t-τ_m) exp[-j(ω(t-τ_m)+θ)], and this output signal is multiplied in a multiplication module 69-m by a complex number a_m exp(jθ_m) (complex path gain for tap m). The numbers a_m and θ_m are selected by the algorithm module 67. A sum signal ##EQU4## is formed by the sum module 57 and is provided at the second input terminal of the difference module 55, as noted before. The algorithm module 67 selects the parameters a_m, θ_m and τ_m for the tapped delay line 65 and the multiplication modules 69-m according to any of a plurality of selected algorithms described below. The tap weighting coefficients a_m exp(jθ_m) are adjusted so that s_M (t;N) resembles the total multipath signal as closely as possible, based upon a criterion associated with the algorithm chosen.

One embodiment 71 of the invention is illustrated in FIG. 7. An incoming signal s(t) in Eq. (1) is passed through a BPF 73 (optional) and through an LPF or BPF 74 (optional), and the resulting baseband signal s_bb (t) is received by the first input terminals of each of a sequence of signal multipliers 77-m (m=1, 2, . . . , N; N≧2). The second input terminals of the signal multiplier 77-m receives a time-delayed signal p(t-τ_m) from a tapped signal delay line 79, which is described below. The individual product signals s_bb (t)p(t-τ_m) are received by a sequence of signal integrators 81-m and integrated with respect to time over a selected time interval [0,T]. The integrated signals are passed through signal amplifiers 83-m (optional) and are received by N input terminals of a code tracking module 85 that selects a tap weighting coefficient a_m exp(jθ_m) and multiplies this coefficient by the input signal s_bb (t)p(t-τ_m) received at the mth input terminal. These product signals are processed by the code tracking module 85 as illustrated in the four preferred embodiments discussed below. The code tracking module 85 controls a code generator module 87 that issues the replica signal p(t). The replica signal p(t) is then passed to the tapped time delay line 79, which forms a sequence of time delayed copies p(t-τ_m) of the input signal and issues these time delayed signals for use in the signal multipliers 77-m as shown. The code tracking module 85 and code generator module 87 rely on algorithms described below.

FIGS. 8A, 8B, 9A, 9B, 10A and 10B illustrate use of an error function e(τ; τ₀ ;τ_d ;r) defined by

e(τ-τ₀ ;τ_d ;r)=|R(τ-τ₀ -τ_d)|^r -|R(τ-τ₀ +τ_d)|^r, (10)

where τ_d is a time value that ranges over the correlator spacing (0≦τ_d ≦T_c) and r is a selected positive number (e.g., r=0.5, 1.0, 1.5, 2.0, .sqroot.π, or any other rational or irrational number). FIGS. 8A, 9A and 10A illustrate three representative shapes for the correlation function magnitude C(τ)=|R(τ)|, and FIGS. 8B, 9B and 10B illustrate the corresponding correlation error functions e(τ-τ₀ ;τ_d ;r) with the choice r=1.

If the time of arrival τ₀ is not initially known, this quantity may be determined by designating τ₀ to be the time value (by interpolation or otherwise) at which the error function e(τ-τ₀ ;τ_d ;r) changes sign. It is possible, although unlikely, that two or more distinct time values exist at which the error function e(τ-τ₀ ;τ_d ;r) changes sign.

If the correlation function magnitude C(τ)=|R(τ)| is nearly symmetric around a time value τ₀, as in FIG. 8A, the error function e(τ-τ₀ ;τ_d ;r) will resemble the function shown in FIG. 8B for all time values τ-τ₀ in the domain of the error function. This condition may be taken to indicate that no multipath signals are present or that the multipath signals are severely attenuated when they arrive at the apparatus 71 in FIG. 7.

I. A first approach that may be used by the algorithm modules 67 and 85 in FIGS. 6 and 7 relies upon a combination of narrow and wide sample spacing of the correlation function. A first pair of samples has narrow sample spacing, typically τ_d between 0.01 T_c and 0.2 T_c, and the second pair of samples has wide sample spacing, typically τ_d between 0.4 T_c and T_c. The receiver estimates the time delay of the signal s(t) using the first pair of correlation signal samples, and the second pair of samples is used to correct the errors introduced by the presence of multipath signals.

FIGS. 11 and 12 exhibit results from a simulation of multipath signal errors in a representative receiver, such as a global positioning system receiver, wherein the abscissa represents the actual signal error experienced by a receiver that uses correlation function sample spacing τ_d (narrow)=0.1 T_c and τ_d (narrow)=0.2T_c, respectively. The ordinate for each figure represents the measurable midpoint error difference Δτ_w,n =Δτ_w -Δτ_n, where the wide sample spacing for each of these figures satisfies τ_d (wide)=0.4T_c. The abscissa for each of these Figures represents the midpoint error Δτ_n with a narrow sample spacing (τ_d =0.1T_c in FIG. 11, τ_d =0.2T_c in FIG. 12) resulting from the simulation. Otherwise stated, the errors shown on the ordinates in FIGS. 11 and 12 are measured by the wide correlator when the receiver is locked to the narrow correlator. FIGS. 11 and 12 indicate that measurable midpoint error differences Δτ_w,n are strongly correlated with the errors Δτ_n produced by the correlator using narrow sample spacing.

Least pth power analysis (e.g., with p=2) can used here to determine a line L of the form y=a x+b (y=Δτ_n, x=Δτ_w -Δτ_n) for each of FIGS. 11 and 12 that most closely fits the simulation data on each of these figures. This analysis minimizes the sum

ε_p (a,b)=Σ_i |a x_i +b-y_i |^p, (11)

with respect to choices of the parameters a and b, for a selected number p>0. Here, (x_i, y_i) is one of the simulation data coordinate pairs in FIGS. 11 or 12. The choice p=2 reproduces the well known least mean squares analysis, with well known expressions for the parameters a and b. Alternatively, any other suitable optimization procedure can be used here, with a different choice of p. A particular best fit to the simulation data shown in FIGS. 11 and 12, is represented by a line L of the form

Δτ_n =a(Δτ_w -Δτ_n)+b,(12)

where Δτ_n and Δτ_w -Δτ_n are the variables associated with the abscissa and ordinate values, respectively, for the line L obtained from FIGS. 11 or 12. The correlation parameters a and b for the data illustrated in FIGS. 11 and 12 and the choice p=2 have the approximate values a(τ_d =0.1T_c, 0.4T_c)=-0.1901, b(τ_d =0.1T_c, 0.4T_c)≈0, a(τ_d =0.2T_c, 0.4T_c)=-0.6803, b(τ_d =0.2T_c, 0.4T_c)≈0, indicating a strong correlation between the correlation function midpoint errors with the measurable quantity Δτ_w -Δτ_n in each instance. For each different pair of choices for narrow spacing and wide spacing, a new optimal parameter pair (a,b) will be found. For the first algorithm used in the invention, the relationship of Eq. (12) is used as indicated below to reduce the error in time of arrival of the direct signal and to thereby reduce the multipath signal error in the composite signal.

Let (Δτ_n, Δτ_w -Δτ_n) be a coordinate pair of error values shown in FIGS. 11 and 12 for the narrow and wide spacing samples of the correlation function R(τ). With reference to FIG. 5, Δτ_n and Δτ_w are the differences between the "true" position τ=τ_peak of the correlation function peak and the estimated peak positions, obtained by using the respective estimates

τ_n,avg =(τ₁,n +τ₂,n)/2, (13)

τ_w,avg =(τ₁,w +τ₂,w)/2; (14)

that is,

Δτ_n =τ_peak -τ_n,avg, (15)

Δτ_w =τ_peak -τ_w,avg. (16)

It is assumed here that a representative signal receiver has been tested sufficiently to obtain a collection of measured error value pairs (x_i, y_i)=(Δτ_n,i, Δτ_w,i -Δτ_n,i) as shown in FIGS. 11 and 12, for a selected choice of narrow correlation sample spacing τ_n and wide correlation sample spacing τ_w and that the optimal parameters a and b for Eq. (11) or Eq. (12) have been determined.

In a particular situation, it is impossible to obtain either of the error values Δτ_n or Δτ_w because the location τ=τ_peak is not known. However, the difference Δτ_w -Δτ_n can be measured and Eq. (12) can be used to estimate the error associated with by the narrow correlator midpoint value. Ideally, the relation

τ_n,avg -τ_peak -a(Δτ_w -Δτ_n)-b=0 (17)

should be true, where the parameters a and b are determined from simulations such as those shown in FIGS. 11 and 12. Equation (17) can then be rearranged to provide an improved estimate for the unknown variable τ_peak, namely

τ_peak,est =τ_n -a(Δτ_w -Δτ_n)-b(18)

With this more accurate estimate τ_peak,est in hand, the arrival time of a multipath signal peak can be estimated and this unwanted signal can be removed from the composite signal s(t) in Eq. (1).

II. A second approach uses a least squares estimation of the multipath signal strengths, using multiple samples {x(τ_m)}_m (preferably at least 10-30 such samples) of the correlation signal values. Each correlation function value x_k can be expressed as a sum ##EQU5## where e_k is a measurement error. Here, τ'_k is drawn from a set of correlator time value samples and τ_m is drawn from a set (of estimated set size J) for times of arrival at the receiver of direct and multipath signals. The quantities R(τ_m -τ_k ') have measurable, and thus known, values, which may be determined using Eq. (6). Equation (19) represents K relations (k=1, 2, . . . , K) for the unknown complex variables b_m =a_m exp{j(θ-θ_m)}, sometimes referred to as the complex path gains. Values for the complex path gains b_m are chosen that minimize the sum of the pth powers of the residuals or errors ##EQU6## with p=2 preferred. The choice for the N×1 vector b with entries b_m that produces this minimum for p=2 is given by

b=(R^tr R)-1 R^tr x, (21)

where R is a J×N matrix with entries R_mk =R(τ_m -τ'_k) (m=1, . . . , J; k=1, . . . , N) and x is a J×1 vector whose entries are the measurable quantities x_j. This algorithm is relatively simple to apply and can identify situations where the correlation function has two or more local maxima and where the narrow sample spacing correlation values may straddle a false peak for this function. A difference signal ##EQU7## is then formed as an estimate of a multipath-free transmitted signal received at the receiver. The earliest time, t=τ₁, is interpreted as the time of arrival of the direct signal.

III. In a least squares or least pth power approach, the times of arrival or path delays for the direct signal and of the multipath signal(s) are best determined separately. A first approach for estimating these path delays uses slope transition points to estimate the multipath parameters. The slope transition approach discussed here, and the cepstrum processing approach discussed below, allows determination of the time value τ=τ_n corresponding to arrival of the direct signal and of each of the multipath signals that contribute to the received composite signal s(t) in Eq. (1), before the complex gain parameters a_m exp(jθ_m) are determined. Once the full set of these time values {τ_m }_m is determined, determination of the complex gain parameters becomes a tractable linear problem, if a least squares method is used for such determination. Thus, use of the slope transition approach or of the cepstrum processing approach decomposes the problem of determination of the parameter sets {τ_m }_m, {a_m }_m, and {θ_m }_m into two independent sub-problems, and one of these sub-problems merely requires solution of an array of linear equations.

Each slope transition point for a correlation function R(τ) corresponds to a unique, somewhat sharper slope transition point for the error function e(τ-τ₀ ;τ_d ;r) for r≧1. The slope transition points t_s for the correlation function R(τ) can be found by visual inspection of the graph of the error function e(τ-τ₀ ; τ_d ;r), or the slope transition points may be determined numerically by a microprocessor connected to a memory module that stores the quantities R(τ-τ₀ ±τ_d). A slope transition point need not be a local maximum of the correlation function R(τ).

FIG. 4 illustrates a representative correlation function R(τ) that might be generated for the composite signal s(t) of Eq. (1). Here, the correlation function has a maximum at τ-τ₀ =0 as before and manifests an abrupt change in slope at τ=τ₀, τ=τ₀ -0.7 T_c and at τ=τ₀ +0.35 T_c, and the slope of the correlation function magnitude |R(τ)| for τ<τ₀ +t_s and for τ>τ₀ +t_s differ substantially from each other. Each time point τ=τ₀ +t_s in the domain of the correlation function where the function slope changes abruptly (e.g., where the two one-sided derivatives (dR/dτ)_τ0+t_s-0 and (dR/dτ)_τ0+t_s+0 are not equal, or where the magnitude of the difference of these derivatives exceeds a selected threshold number) is referred to herein as a "slope transition point". Each slope transition point τ=τ₀ +t_s potentially corresponds to a time of arrival τ=τ_s =τ₀ +t_s of a distinct multipath contribution to the signal S_M (t;N) in Eq. (1). By determining the number of, and the time values of each of, the slope transition points, one can identify the corresponding times τ_s =τ₀ +t_s and the integer N to be used in Eqs. (1) and (2) to construct the approximate function S_M (t;N) in Eq. (1) representing the add-on signal that arrives at the embodiment 51 in FIG. 6.

Determination of the slope transition points t_s and/or the time of arrival τ₀ of the direct signal for the correlation signal R(τ) in Eq. (6) can be implemented in several ways. One approach examines the function e(τ-τ₀ ;τ_d ;r) given in Eq. (10): the time values ts correspond to the zero crossing points t_s at which the function e(τ-τ₀ ;τ_d ;r) changes sign or abruptly changes value, where t_s ranges over a selected sequence of time values in the domain -T_c ≦t_s ≦T_c and r is a selected positive rational or irrational number (e.g., r=0.5, 1.0, 1.5, 2.0, .sqroot.17 or π).

IV. Another approach for estimation of path delays uses cepstrum processing, wherein the logarithm of the Fourier transform of the incident signal s(t) is formed and manipulated as indicated in FIG. 14 to identify certain parameters (here, time values τ for peaks, as in FIGS. 13A and 13B) of any multipath signals that are present. Assume, for purposes of illustration, that the signal s(t) includes a direct signal s₀ (t) and a single multipath signal contribution s₁ (t), as in Eq. (1) with N=1, viz.

s(t)=b₀ p(t-τ₀)+b₁ p(t-τ₁). (23)

b_m =a_m exp(j(θ_m -θ*)) (24)

The Fourier transform of Eq. (23) becomes

S(f)=Φ(s(t))={b₀ exp(-j2πfτ₀)+b₁ exp(-j2πfτ₁)}P(f), (25)

P(f)=Φ(p(t)), (26)

which yields a relation

Φ(s(t))=b₀ exp(-j 2πfτ₀) P(f) [α1 exp(j(2πf(τ₁ -τ₀))+1]-1, (27)

α1=b₁ /b₀ (assumed<<1 in magnitude). (28)

In cepstrum processing, one forms the logarithm (natural, Naperian or other) of each side of Eq. (27), which yields ##EQU8## The inverse Fourier transform of Eq. (29) is then formed, which yields ##EQU9## where Φ-1 {g(f)} denotes the inverse Fourier transform of the function g(f) and δ(w) is a one-dimensional delta function. The first term on the right hand side of Eq. (30) represents the contribution of the direct signal, multiplied by a multiplier b₀ exp[-j2πfτ₀ ]. One effect of presence of a multipath signal with strength (proportional to α₁) is the appearance of an infinite series of approximately equally spaced delta functions centered at the time values t=m(τ₁ -τ₀)(m=0, 1, 2, . . . ). This infinite series can be used to estimate the path delays τ_m associated with arrival of one or more multipath signals.

After the path delays τ_m are determined using the slope transition method or the cepstrum processing method, a least squares estimation method (second approach) can be used to determine the complex path gain parameters b_m in Eq. (22) and to determine an estimate of the multipath signal. This estimate can be subtracted from the received signal to produce a signal that is substantially multipath-free. The signal receiver can then estimate the arrival time of the multipath-free signal, with reduced multipath errors therein.

V. Another, more direct, approach for determining the path delays, or arrival times for the direct and multipath signals, begins with a selected grid or sequence of time values {τ_g }_g (g=1, 2, . . . , G) along the interval -T_c ≦t≦T_c and minimizes the quantity ε_p in Eq. (20) for this sequence, where the time values τ_m in Eq. (20) are now drawn from this grid of values {τ_g }_g. This minimization produces an G×1 vector b of complex gain values, defined by Eq. (21). This approach seeks to identify the largest values of the complex path gain magnitudes |b_g |, according to some algorithm or other identification procedure.

One method used with this approach arranges the magnitudes |b_g | of the complex path gains in decreasing order of the magnitudes and chooses a subset of H complex path gains b_h with the largest magnitudes (h=1, 2, . . . , H), where H is a selected positive integer. The time values τ_h corresponding to these complex path gains b_h are identified as the times of arrival of the direct signal (largest path gain magnitude) and of the multipath signals.

A second method here again arranges the magnitudes |b_g | of the complex path gains in decreasing order of the magnitudes and selects all complex path gains b_h satisfying the relation

|b_h |≧δ, (31)

where δ is a selected threshold number, with the largest path gain magnitude corresponding to the time of arrival of the direct signal.

A third method here arranges the magnitudes |b_g | of the complex path gains in decreasing order of the magnitudes, written in that order as

|b_g1 |≧|b_g2 |≧|b_g3 |≧ . . . ,(32)

and defines

Δb_k =|b_gk |-|b_g(k+1) |≧0, (33)

m(b)=min_k (Δb_k), (34)

M(b)=max_k (Δb_k). (35)

One then seeks the largest number μ(b) lying in the range m(b)≦μ(b)≦M(b) for which

Δb_k ≧μ(b) (k=1, 2, . . . , K) (36)

for a consecutive sequence of K magnitude differences Δb_k and chooses the time values τ_g1, τ_g2, . . . , τ_gK as the times of arrival of the direct and multipath signals. The integer K is not chosen a priori but is determined by the requirement associated with Eq. (36). Equation (36) is always satisfied by at least one value, namely μ(b)=Δb₁ =|b_g1 |-|b_g2 |. Any of these three methods, and others as well, can be used to estimate a reasonable number of arrival times for the direct and multipath signals.

If the grid or sequence of time values {τ_g }_g is a uniformly spaced grid with a nearest neighbor spacing of Δτ≈2T_c /G, the maximum error in the direct signal time of arrival is estimated as Δτ/2. One can choose the grid spacing Δτ so that this maximum error is no larger than an acceptable amount.

As an enhancement, this uniform grid, centered at an estimated arrival time t=τ_g1, may be replaced by a uniform grid g(j;J;Δt) with about the same grid spacing, with grid markers τ_g1 (j) that are each translated by a distance (j/J)(T_c /G) to the right (or to the left) of the grid marker of interest, t=τ_g1, with j=0, 1, . . . , J-1. The time of arrival of the direct signal is then the time τ_g1 (j') that corresponds to the integer value j' for which τ_g1 (j')=max₁≦j≦J-1 τ_g1 (j). With this alternative adopted the maximum error in the direct signal time of arrival is reduced further to Δτ/J. This grid translation process corresponds to imposing a finer grid, with separation distance T_c /GJ, in a neighborhood of the grid marker of interest, t=τ_g1, and repeating the preceding computations to obtain a better estimate of the direct signal arrival time.

FIG. 15 is a flow chart illustrating a first embodiment of the invention, in which wide and narrow correlator function sampling is used to identify the time value corresponding to a correlator peak value. In step 121, a carrier signal with selected frequency f=ω/2π, modulated with a selected shape signal p(t) at a selected frequency f_c =1/T_c, is transmitted. In step 123, the transmitted signal is received as a composite signal, containing the direct signal and possible multipath signals, at a receiver. In step 125, a correlation function R(τ) is formed as in Eq. (6).

In steps 131 and 133, time values τ₁,w, τ₂,w, τ₁,n and τ₂,n are determined that satisfy the constraints |R(τ₁,w)|=|R(τ₂,w )| and |R(τ₁,n)|=|R(τ₂,n | with Δτ_w,sp =τ₂,w -τ₁,w and Δτ_n,sp =τ₂,n -τ₁,n, with 0.4T_c ≦Δτ_w,sp ≦T_c and 0.01T_c ≦Δτ_n,sp ≦0.2T_c. In step 135, a collection of simulated pairs of values {(Δτ_w,k -Δτ_n,k, Δτ_n,k)}_k is provided. In step 137, real parameters a and b are determined that provide an optimal fit to the equation Δτ_n ≈a(Δτ_w -Δτ_n)+b for this collection of simulated pairs of values. In step 139, a time of arrival, τ_peak, of the direct signal is estimated by the equation τ_peak =(τ₁,n +τ₂,n)/2-a(Δτ_w -Δτ_n)-b.

FIG. 16 is a flow chart illustrating a second embodiment of the invention, using minimization of the difference between the received signal and a sum of replicas of the shape signal p(t) with unknown weights. In FIG. 16, steps 121, 123 and 125 are as in FIG. 15. In step 141, estimates of the times of arrival τ_m (m=1, 2, . . . , N+1; N≧1) of the direct signal and one or more multipath signals are obtained independently, using, for example, a method such as identification of slope transition points or Cepstrum processing, as discussed above. In step 143, a sequence of scalar equations is formed as in Eq. (19). In step 145, a pth power difference, given in Eq. (20), is minimized with respect to choice of the parameters b_m =a_m exp[j(θ-θ_m)]. In step 147, a difference signal Δs(t) is formed as in Eq. (22), as an estimate of a multipath-free signal received at the receiver.

FIG. 17 is a flow chart illustrating a third embodiment of the invention, which uses a pre-selected grid of time values {τ_g }_g. In FIG. 17, steps 121, 123 and 125 are as in FIG. 15. In step 151, a grid of time values {t_g }_g (g=1, 2, . . . , G) is chosen along the time interval -T_c ≦t≦T_c. In step 153, a sequence of scalar equations is formed as in Eq. (19), with the time values tm and tk being drawn from the grid {τ_g }_g. In step 155, the parameters b_g are chosen to minimize the pth power difference given in Eq. (20). In step 157, the magnitudes |b_g | of these determined parameters are arranged in a sequence {τ_gk }_gk, of decreasing magnitudes with |b_g1 |≧|b_g2 |≧|b_g3 |≧ . . . . In step 159, the first N+1 signals in the sequence {τ_gk }_gk are selected and a difference signal Δs(t) given in the form of Eq. (22) is adopted as an estimate of a multipath-free signal received at the receiver.

Claims

1. A method for compensating for the presence of one or more multipath signals in an electromagnetic signal received at a receiver, the method comprising the steps of:

receiving a signal s(t) at a receiver that is the result of a transmitted message signal with frequency ω and having a shape signal p(t) with frequency f_c =1/t_c ;

selecting a grid of values {τ_g }_g (g=1, 2, . . . , G) of the time variable t lying in the interval -t_c ≦t≦T_c ;

selecting a set of sampling times (τ'_j)_j (j=1, . . . , J; J a selected integer≧2) for a correlation function r(τ), and determining the correlation function values r(τ'_j)=x_j, where the correlation function r(τ) is defined by ##EQU10## and t is a selected time satisfying T>>t_c ; choosing complex numbers b_g so that a sum ##EQU11## is minimized, where q is a selected positive number; arranging the magnitudes |b_g | of the complex numbers b_g in a sequence {|b_gk |}_gk in decreasing order of magnitude so that |b_g1 |≧|b_g2 |≧|b_g3 |≧ . . . , and choosing H members of this sequence, where H is a selected positive integer;

designating the time t=τ_g1 as the time of arrival of the direct signal and the times t=τ_gk (k=2, . . . , H) as the times of arrival of multipath signals; and

interpreting a difference signal ##EQU12## as a signal from which at least one multipath signal has been removed.

2. The method of claim 1, wherein said step of choosing said H members of said sequence {|b_gk |}_gk comprises the steps of:

selecting a positive integer H; and

choosing the largest H members of said sequence {|b_gk |}_gk.

3. The method of claim 1, wherein said step of choosing said H members of said sequence {|b_gk |}_gk comprises the step of choosing all members of said sequence {|b_gk |}_gk that satisfy the relation |b_gk |≧δ, where δ is a selected positive number.

4. The method of claim 1, wherein said step of choosing said H members of said sequence {|b_gk |}_gk comprises the steps of:

forming said sequence of values {Δb_gk }_gk, where Δb_gk =|b_gk |-|b_g(k+1) |;

determining a largest value μ(b) for which Δb_gk ≧μ(b) for all numbers gk in the range 1≦gk≦gK for some positive integer k; and

selecting said integer J=K and selecting said sequence {Δb_gk }_gk (k=1, . . . , k) as said J members of said sequence {Δb_gk }_gk.

5. A method for compensating for the presence of one or more multipath signals in an electromagnetic signal received at a receiver, the method comprising the steps of:

receiving a signal s(t) at a receiver that is the result of a transmitted message signal with frequency ω and having a shape signal p(t) with frequency f_c =1/t_c ;

forming a correlation function r(τ) defined by ##EQU13## where t is a selected time satisfying T>>t_c ; determining a first pair of distinct time values τ=τ₁,w and τ=τ₂,w which produce equal values of the correlation function magnitude |r(t)|, using a difference Δτ_w,sp =τ₂,w -τ₁,w in the range 0.4 t_c ≦Δτ_w,sp ≦T_c, and computing τ_w,avg =(τ₁,w +τ₂,w)/2;

determining a second pair of distinct time values τ=τ₁,n and τ=τ₂,n which produce equal values of the correlation function magnitude |r(t)|, using a difference Δτ_n,sp =τ₂,n -τ₁,n in the range 0.01 t_c ≦Δτ_n,sp ≦0.2 t_c, and computing τ_w,avg =(τ₁,w +τ₂,w)/2, where Δτ_n =τ_n,avg -τ_peak and Δτ_w =τ_w,avg -τ_peak and τ_peak is the value of τ, to be determined, for the peak of the correlation function magnitude;

providing a collection {(Δτ_w,k,sim -Δτ_n,k,sim, Δτ_n,k,sim)}_k of a selected number k of coordinate pairs of simulated values of the variables Δτ_w -Δτ_n and Δτ_n, and determining real parameters a and b that provide an optimal approximation to this collection of coordinate pairs by the equation Δτ_n,sim =a(Δτ_w,sim -Δτ_n,sim)+b;

determining the time value τ_peak of the correlation function magnitude by the relation τ_peak =τ_n,avg -a(Δτ_w -Δτ_n)-b; and

interpreting the time value τ_peak as the time a direct signal arrives at the receiver.

6. The method of claim 5, wherein said step of determining said optimal approximation to said collection of coordinate pairs comprises the step of determining said parameters a and b that minimize a sum ##EQU14## where q is a selected positive number.

7. Apparatus for compensating for the presence of one or more multipath signals in an electromagnetic signal received at a receiver, the apparatus comprising:

a receiver that receives a signal s(t) that is the result of a transmitted message signal with frequency ω and having a shape signal p(t) with frequency f_c =1/t_c ;

a signal processor, connected to the receiver, that:

(1) forms a correlation function r(τ) defined by ##EQU15## where t is a selected time satisfying T>>t_c ; (2) determines a sequence of N time values (N a selected integer≧2) {τ_m }_m, each of which is estimated to correspond to a time of arrival of a direct signal or of a multipath signal at the receiver;

(3) selects a sequence of sampling times {τ'_j }_j (j=1, . . . , J; J a selected integer≧2) for the correlation function r(τ) and determines the correlation function values r(τ'_j)=x_j ;

(4) determines complex numbers b_m so that a sum ##EQU16## is minimized, where q is a selected positive number; and (5) interprets a difference signal ##EQU17## as a signal from which at least one multipath signal has been removed.

8. The apparatus of claim 7, wherein said signal processor determines said sequence of time values {τ_m }_m by:

forming a logarithm of a fourier transform, log{Φ(s(t))} of said received signal;

forming an inverse fourier transform X(t)=Φ^{-1 {log {Φ(s(t))}}; and}

identifying a time value t=τ₁ for which the magnitude |X(t)| has a peak, as a time at which a multipath signal arrives at said receiver.

9. The apparatus of claim 7, wherein said signal processor determines said sequence of time values {τ_m }_m by (i) identifying at least two slope transition points for said correlation function r(τ) and (ii) for each slope transition point identified, identifying at least one of said time values τ_m with this slope transition point.

10. The apparatus of claim 9, wherein said signal processor identifies said slope transition point by:

forming an error function e(τ;τ₀ ;τ₁ ;q) defined by

e(τ-τ₀ ;τ_d ;r)=|r(τ-τ₀ -τ_d)|^r -|r(τ-τ₀ +τ_d)|^r,

where τ_d is a time value that ranges over a selected sequence of distinct time values {t'_u }_u in the range -t_c ≦t'_u ≦T_c, and r is a selected positive number;

determining at least one value τ=τ' for which the error function e(τ-τ₀ ;τ_d ;r) has a slope transition point; and

interpreting the time value τ' as a slope transition point for the correlation function r(τ).

11. The apparatus of claim 9, wherein said signal processor identifies said slope transition point by:

determining the one-sided derivatives r'(τ)_{+ =(dR/dτ)τ+0 and r'(τ)- =(dR/dτ)τ-0 of said correlation function r(τ) at a selected sequence of points τ=τk (k=1, 2, . . . ) in an interval -tc ≦τk ≦Tc ; and}

when two one-sided derivatives r'(τ)_{+ and r'(τ)- satisfy the relation |r'(τk)+ -r'(τk)- |≧δ for at least two of the points τk in this sequence, where δ is a selected positive constant, designating these points δk as slope transition points.}

12. Apparatus for compensating for the presence of one or more multipath signals in an electromagnetic signal received at a receiver, the apparatus comprising:

a receiver that receives a signal s(t) that is the result of a transmitted message signal with frequency ω and having a shape signal p(t) with frequency f_c =1/t_c ;

a signal processor, connected to the receiver, that:

(1) forms a correlation function r(τ) defined by ##EQU18## where t is a selected time satisfying T>>t_c ; (2) selects a sequence of values {τ_g }_g of the time variable τ lying in the interval -t_c ≦τ≦T_c,

(3) selects a sequence of sampling times {τ'_j }_j (j=1, . . . , J; J a selected integer≧2) for the correlation function r(τ) and determines the correlation function values r(τ'_j)=x_j ;

(4) determines complex numbers b_g so that a sum ##EQU19## is minimized, where q is a selected positive number; (5) arranges the magnitudes |b_g | of the complex numbers b_g in a sequence {|b_gk |} in decreasing order of magnitude so that |b_g1 |≧|b_g2 |≧|b_g3 |≧ . . . , and chooses H members of this sequence, where H is a selected positive integer;

(6) designates the time t=τ_g1 as the time of arrival of the direct signal and at least one of the times t=τ_gk (k=2, . . . , H) as the times of arrival of multipath signals; and

(7) interprets a difference signal ##EQU20## as a signal from which at least one multipath signal has been removed.

13. The apparatus of claim 12, wherein said signal processor chooses said H members of said sequence {|b_gk|}_gk by: (1) selecting a positive integer H; and (2) choosing the largest H members of said sequence {|b_gk |}_gk.

14. The apparatus of claim 12, wherein said signal processor chooses said H members of said sequence {|b_gk |}_gk by choosing all members of said sequence {|b_gk |}_gk that satisfy the relation |b_gk |≧δ, where δ is a selected positive number.

15. The apparatus of claim 12, wherein said signal processor chooses said H members of said sequence {|b_gk |}_gk by:

forming a sequence of values {Δb_gk }_gk, where Δb_gk =|b_gk |-|b_g(k+1) |;

determining a largest value μ(b) for which Δb_gk ≧μ(b) for all numbers gk in the range 1≦gk≦gK for some positive integer k; and

selecting said integer H=K and selecting the sequence {Δb_gk }_gk (k=1, . . . , k) as said H members of said sequence {Δb_gk }_gk.

16. Apparatus for compensating for the presence of one or more multipath signals in an electromagnetic signal received at a receiver, the apparatus comprising:

a receiver that receives a signal s(t) that is the result of a transmitted message signal with frequency ω and having a shape signal p(t) with frequency f_c =1/t_c ;

a signal processor, connected to the receiver, that:

(1) forms a correlation function r(τ) defined by ##EQU21## where t is a selected time satisfying T>>t_c ; (2) determines a first pair of distinct time values τ=τ₁,w and τ=τ₂,w which produce equal values of the correlation function magnitude |r(t)|, using a difference Δτ_w,sp =τ₂,w -τ₁,w in the range 0.4 t_c ≦Δτ_w,sp ≦T_c, and computes τ_w,avg =(τ₁,w +τ₂,w)/2;

(3) determines a second pair of distinct time values τ=τ₁,n and τ=τ₂,n which produce equal values of the correlation function magnitude |r(t)|, using a difference Δτ_n,sp =τ₂,n -τ₁,n in the range 0.01 t_c ≦Δτ_n,sp ≦0.2 t_c, and computes τ_w,avg =(τ₁,w +τ₂,w)/2, where Δτ_n =τ_n,avg -τ_peak and Δτ_w =τ_w,avg -τ_peak and τ_peak is the value of τ, to be determined, for the peak of the correlation function magnitude;

(4) provides a collection {(Δτ_w,k,sim -Δτ_n,k,sim, Δτ_n,k,sim)}_k of a selected number k of coordinate pairs of simulated values of the variables Δτ_w -Δτ_n and Δτ_n, and determines real parameters a and b that provide an optimal approximation to this collection of coordinate pairs by the equation Δτ_n,sim =a(Δτ_w,sim -Δτ_n,sim)+b;

(5) determines the time value τ_peak of the correlation function magnitude by the relation τ_peak =τ_n,avg -a(Δτ_w -Δτ_n)-b; and

(6) interprets the time value τ_peak as the time a direct signal arrives at the receiver.

17. The method of claim 16, wherein said signal processor determines said optimal approximation to said collection of coordinate pairs by determining said parameters a and b that minimize a sum ##EQU22## where q is a selected positive number.

18. A method for compensating for the presence of one or more multipath signals in an electromagnetic signal received at a receiver, the method comprising the steps of:

receiving a signal s(t) at a receiver that is the result of a transmitted message signal with frequency ω and having a shape signal p(t) with frequency f_c =1/t_c ;

forming a correlation function r(τ) defined by ##EQU23## where t is a selected time satisfying T>>t_c ; determining a sequence of N time values (N a selected integer≧2) {τ_m }_m, each of which corresponds to a time of arrival, lying in the interval -t_c ≦t≦T_c, of a direct signal or of a multipath signal at the receiver, wherein the time values τ_m are determined by (i) forming the logarithm of a fourier transform log{Φ(s(t))} of the received signal s(t), (ii) forming an inverse fourier transform X(t)=Φ^{-1 {log {Φ(s(t))}}, and (iii) identifying at least one value t=τ_m (m=1, 2, . . . , N) for which the magnitude |X(t)| has a peak;}

selecting a set of sampling times {τ'_k }_k (k=1, . . . , k; k a selected integer≧2) for the correlation function r(τ), and determining the correlation function values r(τ'_k)=x_k ;

choosing complex numbers b_m so that a sum ##EQU24## is minimized, where q is a selected positive number; and interpreting a difference signal ##EQU25## as a signal from which at least one multipath signal has been removed.

19. The method of claim 18, further comprising the step of identifying at least one of said time values τ_m with a time of arrivel of a direct signal or of a multipath signal.

20. A method for compensating for the presence of one or more multipath signals in an electromagnetic signal received at a receiver, the method comprising the steps of:

receiving a signal s(t) at a receiver that is the result of a transmitted message signal with frequency ω and having a shape signal p(t) with frequency f_c =1/t_c ;

forming a correlation function r(τ) defined by ##EQU26## where t is a selected time satisfying T>>t_c ; determining a sequence of N time values (N a selected integer≧2) {τ_m }_m, each of which corresponds to a time of arrival, lying in the interval -t_c ≦t≦T_c, of a direct signal or of a multipath signal at the receiver, wherein the time values τ_m are determined by (i) identifying at least two slope transition points for the correlation function magnitude |r(τ)|, and (ii) for each slope transition point identified, identifying at least one time value τ_m with each slope transition point;

selecting a set of sampling times {τ'_k }_k (k=1, . . . , k; k a selected integer≧2) for the correlation function r(τ), and determining the correlation function values r(τ'_k)=x_k ;

choosing complex numbers b_m so that a sum ##EQU27## is minimized, where q is a selected positive number; and interpreting a difference signal ##EQU28## as a signal from which at least one multipath signal has been removed.

21. The method of claim 20, further comprising the step of identifying at least one of said time values τ_m with a time of arrivel of a direct signal or of a multipath signal.

22. The method of claim 20, wherein said step of identifying said slope transition point comprises the steps of:

forming an error function e(τ-τ₀ ;τ_d ;r) defined by

e(τ-τ₀ ;τ_d ;r)=|r(τ-τ₀ -τ_d)|^r -|r(τ-τ₀ +τ_d)|^r,

where τ_d is a time value that ranges over a selected sequence of distinct time values {t'_u }_u in the range 0≦t'_u ≦T_c, and r is a selected positive number;

determining at least one value τ=τ' for which the error function e(τ-τ₀ ;τ_d ;r) has a slope transition point; and

interpreting the time value τ' as a slope transition point for said correlation function r(τ).

23. The method of claim 20, wherein said step of identifying said slope transition point comprises the steps of:

determining the one-sided derivatives r'(τ)_{+ =(dR/dτ)τ+0 and r'(τ)- =(dR/dτ)τ-0 of said correlation function r(τ) at a selected sequence of points τ=τk (k=1, 2, . . . ) in an interval -tc ≦τk ≦Tc ; and}

when two one-sided derivatives r'(τ_k)_{+ and r'(τk)- satisfy the relation |r'(τk)+ -r'(τk)- |≧δ for at least two of the points τk in this sequence, where δ is a selected positive constant, designating these points τk as slope transition points.}